A. Joseph Dupre III
Technical Product Manager
Blog

How to Enable BreakingPoint VE Performance Acceleration Using Intel DPDK

October 17, 2016 by A. Joseph Dupre III

DPDK1For the last several releases of BreakingPoint Virtual Edition (VE), Ixia’s engineering team has been hard at work squeezing out ever last ounce of performance capability they can get their hands on. This has led to a considerable increase in the scalability of the Virtual Edition for modeling real-world application traffic, security strikes, and malware.

Application test teams continue to push the performance limits of what common off-the-shelf hardware and virtual network function infrastructure (VNFI) can achieve. To stay ahead of the explosion in growth of performance testing of virtual network infrastructures, our engineering team is leveraging every tool in their toolbox.  Long gone are the days of simply needing to verify just basic connectivity between groups of unrelated virtual machines. Today’s virtual networks are many-layered and have complex interdependencies, yet still need to fulfill the service level agreements (SLAs) traditionally serviced by dedicated hardware.

The industry as a whole has been working on getting as much performance as possible out of VNFI, which has led to several optimization technologies and strategies. One key technology that is making its way into a wide variety of VNFs is the data plane development kit (DPDK) from Intel.

According to Intel, packet processing and throughput is significantly enhanced to the tune of up to 10X improvement compared with systems that have not used the DPDK.  Software engineers can compile their VNFs with the DPDK code base to take advantage of the kit’s libraries for executing code in the Linux user space that provides several optimizations. There are software abstractions for large-scale multi-core systems, manager routines for memory via use of a ring to store free objects in the memory pool, and poll mode drivers that work asynchronously to reduce processing overhead—just to name a few.

Accelerate Your Real-World Test Traffic

Users can now accelerate the performance of BreakingPoint VE in generating application test traffic and security test traffic by taking advantage of Intel DPDK functionality. This new Performance Acceleration feature has some important prerequisites:

  1. The server processor that the virtual blade will be using needs to have single instruction multiple data (SIMD) extensions of SSSE3 or above enabled
  2. At least 8GB of RAM needs to be allocated for the virtual blade
  3. The hypervisor needs to be a version of VMware ESXi 6.0 with a build number of 3029758 or above

DPDK2

Figure 1. Performance Acceleration configuration options on virtual blade

The BreakingPoint VE open virtual architecture appliance (OVA) for the virtual blade can be downloaded from the Ixia Support website and deploys with VMXNET3 interfaces on VMware ESXi to take advantage of the Intel DPDK enhancements supported by newer versions of VMware Hypervisor. Lastly, Ixia engineering recommends using default settings for VMware hypervisor configuration options under the Hypervisor->Configuration->Software->Advance Settings->Net path.

There are a few steps needed to make use of the new Performance Acceleration feature. The first step is to configure (TCP segmentation offload) TSO and (large receive offload) LRO in the VMware ESXi 6.0 hypervisor. Refer to the VMware knowledge bases for great details on how to verify or enable TSO and LRO if they are disabled.

The second step is to navigate to the Device Status page in the BreakingPoint web interface. Each vBlade on the Device Status page of the GUI displays a green slot configuration button at the top-right corner. When you click the slot configuration button, you’ll see a pop-up window offering an option to Enable Performance Acceleration.  Select the Enable Performance Acceleration option and click the Apply button. 

Note that there is a warning regarding the fact that the virtual blade will need to automatically reboot itself to switch modes into the performance optimized code path, which is not the default. The blade will disappear from the Device Status page after a brief period of time and will reappear upon completion of the reboot. If you wish to verify that the change is in effect on a virtual blade, you can again click the green slot configuration button and see the checkbox next to Enable Performance Acceleration and the Apply button will be grayed out.

The final step is to select a Performance Emphasis of Throughput from the dropdown in each Application Sim component you are going to use in a test that makes use of the virtual blades you previously enabled for Performance Acceleration. The default Performance Emphasis is balanced and while you will benefit from performance optimizations applied by the Ixia engineering team over the last several releases, you need to use Throughput Emphasis to signal to the BPS engine that you want the additional benefits of Intel DPDK in a VMXNET3 environment. 

DPDK3             

Figure 2. Performance Emphasis configuration options within an Application Sim component

Performance Acceleration support via Intel DPDK over VMware VMXNET3 network interfaces is now available with the BreakingPoint 8.10 EA release. For more details on the Performance Acceleration feature in BreakingPoint VE please consult the BreakingPoint Virtual Edition Installation Guide and the BreakingPoint User Guide.